Darknet: train and test set in YOLO v3

[jain2anki]

my dataset contains few images (~490 images for 2 classes). so it is fine if don't split the dataset into train and test set and use all the images for training? Does the size of test set affect the training of yolo?

[Rajasekhar06]

Nope, it does not affect training but while calculating the accuracy for your created model then you might need to provide the images which are never shown to the model to check how accurately it is getting predictions while calculating mAP other than that there is a no use of test and train set division

[jain2anki]

and how do we calculate accuracy in yolo? I just know how to detect the object, but is there some functionality to test the accuracy given a validation dataset?

[Rajasekhar06]

yes there is a functionality
./darknet detector map data/YourDataFile.data cfg/yourCfgFile.cfg backup/YourWeightsFile.weights -points 0 -thresh 0.1
in data file you need to mention the path of your test set then it will go through all the images and annotations and compare them to the detected classes make sure these images are not from training set give completely new images to find the real accuracy.

[vishnuvardhan58]

Hello @Rajasekhar06 , I am training with 5 species and gave the command you mentioned above and I am getting the results shown in the below image , what exactly is the meaning of TP ,FP and FN for each class ?